upgrade_dfm() not working on some pre-docid_ v2 objects

[chainsawriot]

Describe the bug

I encounter this problem while working with the useNews (@cbpuschmann & @MarHai, 2020) dataset. As the data is not from me, it could be a data issue (e.g. the dfm is malformed). But displaying the dataset doesn't suggest so.

I can't directly pinpoint what is the actual cause of the error. debug() actually points to dfm_match, exactly this line.

Reproducible code

Please paste minimal code that reproduces the bug. If possible, please upload the data file as .rds.

require(osfr)
require(quanteda)

osf_retrieve_node("uzca3") %>% osf_ls_files(n_max = 1, pattern = "usenews.mediacloud.wm.RData") %>% osf_download(path = ".", progress = TRUE)

load("usenews.mediacloud.wm.RData")

## mediacloud.wordmatrix2019 and mediacloud.wordmatrix2020 are a list of dfms.

mediacloud.wordmatrix2019[[1]] ## looks okay, it was created with v 1.5.2
mediacloud.wordmatrix2020[[1]] ## this one might be corrupted, but looks okay, was created with v 2.1.0 

dfm_lookup(mediacloud.wordmatrix2019[[1]], dictionary(list(Boston = c("boston")))) ## works
dfm_match(mediacloud.wordmatrix2019[[1]], "boston") ## works

dfm_lookup(mediacloud.wordmatrix2020[[1]], dictionary(list(Boston = c("boston")))) ## error
dfm_match(mediacloud.wordmatrix2020[[1]], "boston") ## print problem

Expected behavior

It works for 2019 data but not for 2020 data. 2020 data should work.

System information

R version 4.0.4 (2021-02-15)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 20.04.2 LTS

Matrix products: default
BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.9.0
LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.9.0

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] quanteda_3.0.0 osfr_0.2.8    

loaded via a namespace (and not attached):
 [1] Rcpp_1.0.6         magrittr_2.0.1     stopwords_2.2      tidyselect_1.1.0  
 [5] lattice_0.20-41    R6_2.5.0           rlang_0.4.10       fastmatch_1.1-0   
 [9] httr_1.4.2         dplyr_1.0.2        tools_4.0.4        grid_4.0.4        
[13] ellipsis_0.3.1     RcppParallel_5.0.3 digest_0.6.27      httpcode_0.3.0    
[17] tibble_3.0.4       lifecycle_0.2.0    crayon_1.4.1       Matrix_1.3-2      
[21] purrr_0.3.4        vctrs_0.3.5        fs_1.5.0           curl_4.3          
[25] crul_1.0.0         memoise_1.1.0      glue_1.4.2         stringi_1.5.3     
[29] compiler_4.0.4     pillar_1.4.7       generics_0.1.0     jsonlite_1.7.1    
[33] pkgconfig_2.0.3   

Additional info

Both the CRAN version (2.1.2) and the Github version (3.0.0) have the same issue.

[kbenoit]

Thanks for filing this and making it easy to trace. I've figured out the issue.

@koheiw what's happening here is that our is_pre2 returns FALSE so the upgrade_dfm() is never triggered. However even if we do try to run the upgrade_docvars() from that function on this dfm, it fails. I think we must have introduced the new builder functions and the docid_ etc structure just after releasing v2, so didn't catch this case. Would be great to solve this before we release v3. Maybe change is_pre2 conditions to become is_pre3()?

To make this more easily reproducible (since the original files are several GB each) I've extracted just the offending dfm here:

library("quanteda")
## Package version: 3.0.0
## Unicode version: 10.0
## ICU version: 61.1
## Parallel computing: 12 of 12 threads used.
## See https://quanteda.io for tutorials and examples.
problem_dfmat <- readRDS(url("https://www.dropbox.com/s/p15zhbluiltp36x/problem_dfmat.rds?dl=1"))

meta(problem_dfmat, type = "system")
## $`package-version`
## [1] '2.1.0'
## 
## $`r-version`
## [1] '4.0.2'
## 
## $system
##  sysname  machine     user 
## "Darwin" "x86_64"     "cp" 
## 
## $directory
## [1] "/Users/cp/GDrive/Projekte/useNews"
## 
## $created
## [1] "2020-10-30"

# problem is no docid_
names(quanteda:::get_docvars.dfm(problem_dfmat, user = TRUE, system = TRUE))
##  [1] "stories_id"           "processed_stories_id" "collect_date"        
##  [4] "guid"                 "title"                "publish_date"        
##  [7] "url"                  "language"             "ap_syndicated"       
## [10] "media_id"             "media_name"           "media_url"
docid(problem_dfmat)
## Error in select_docvars(x@docvars, field, user, system, drop): field(s) docid_ not found

^{Created on 2021-03-21 by the reprex package (v1.0.0)}

[kbenoit]

@koheiw it's possible the same issue will also affect corpus and tokens objects, but since we didn't get that object "in the wild" I have not confirmed.

[koheiw]

I think that the DFM was created because we forgot to call as.tokens() in dfm.tokens(). I fixed that recently.

[MarHai]

Thanks for the report @chainsawriot and the quick help, @kbenoit and @koheiw. Anything we (as the dataset authors) could do as of now? Could we, for example, trigger any update mechanisms manually to update the datasets?

[kbenoit]

@MarHai I'd suggest

mediacloud.wordmatrix2020 <- lapply(mediacloud.wordmatrix2020, as.dfm)

to update your objects.

@koheiw in #2098, now merged, I noticed that as.dfm() (via upgrade_dfm()) does not update the system meta. We should change that, no? Below it exits before updating the rest.

quanteda/R/object-builder.R

Lines 52 to 58 in ffb814a

	upgrade_dfm <- function(x) {
	if (!is_pre2(x)) return(x)
	attrs <- attributes(x)
	if ("meta" %in% names(attrs)) {
	x@docvars <- upgrade_docvars(attrs$docvars, rownames(x))
	return(x)
	}

[koheiw]

There is only one version of the system meta, so it should be valid if an object has it. ie ("meta" %in% names(attrs)) == TRUE

[kbenoit]

What I meant was we don't update it for the version that performed the upgrade.

> problem_dfmat <- readRDS(url("https://www.dropbox.com/s/p15zhbluiltp36x/problem_dfmat.rds?dl=1"))
> problem_dfmat2 <- as.dfm(problem_dfmat)
> meta(problem_dfmat2, type = "system")
$`package-version`
[1] ‘2.1.0’

$`r-version`
[1] ‘4.0.2’

$system
 sysname  machine     user 
"Darwin" "x86_64"     "cp" 

$directory
[1] "/Users/cp/GDrive/Projekte/useNews"

$created
[1] "2020-10-30"

[koheiw]

I see. I think it is OK without updating the version, because the object structure is of v2.1 (the same in v3.0).

[MarHai]

For the sake of completeness, @chainsawriot, the respective objects in the OSF repository of useNews are now updated as well.